Is Ideogram the Best AI Image Generator?
Let’s look at prompt coherence, text rendering, and image quality
When Ideogram first launched last September, I was really excited because they were the first to solve the text-rendering problem. I made a bunch of experiments, posted a ton content on Instagram, and then sort of lost touch as all the other AI tool advancements rolled out. As Midjourney and Dalle-3 made huge advancements, I had demoted Ideogram to an ‘also-ran’, or a clever toy, but not a serious tool. However, their latest update this week puts them squarely in front runner territory. They have also received a new round of funding from Andreeson Horowitz who have invested another $60 million.
So what makes Ideogram so good? First, their text rendering ability is probably the best out of all the tools, and you’ll get an accurate output most of the time. Sure it won’t be perfect every time and you’ll need to re-roll, but it’s better than the others. Next, they have a ‘prompt-helper’, similar to Dalle-3, that can take your suggestions and turn it into a more effective prompt. But the major advantage is improved prompt coherence, where you can provide simple details and Ideogram can expand those and produce amazing results. At least this is what I’ve been reading and watching throughout the AI-verse this week, but I wanted to put it to a test. I’m going to try and run the same eight prompts through Ideogram, Midjourney and Google Test Kitchen and try to evaluate them on image quality, text rendering, and prompt coherence. I’ll allow a maximum of 2 extra re-rolls, just to give everyone a chance. For the sake of consistency, the images will always be in the same order: Ideogram on the left, Midjourney in the Middle, and Google Test Kitchen on the right. Occasionally, I tweaked the syntax in one tool or another, but the main idea was the same. Ideogram has a magic prompt helper that takes your prompt and extrapolates more complex prompts like this, but I won’t show all of them.
A creative photo of a penguin expertly riding a snowboard, with its legs tucked neatly beneath it. The penguin holds a bold, black banner in its beak, which reads “Winter is Forever.” The image is captured in a snowy landscape, with tall trees and a frosty blue sky. The overall feel of the photo is playful and whimsical, with a touch of winter wonderland charm.
Prompt #1: A Penguin Riding a Snowboard, holding a banner that reads “Winter is Forever”.
While all three penguins are charming, Midjourney is out for that extra ‘is’. I think that Google’s image quality is the best, but it’s a sign, not a banner, so the winner is Ideogram, even if all the words don’t fit in the banner.
Prompt #2 A ceramic mug with text that reads “Coffee is a Human Right”, steaming coffee.
While these are all excellent, Ideogram’s is cast in shadow, Midjourney’s is overly art directed and Google test Kitchen’s has an awkward mix of fonts, but is still the best. Round #2 to Google.
Prompt #3 A craft beer can with an illustrated Bigfoot and text that reads “Sasquatch IPA”.
All three generators struggled with this prompt and I broke my own rules and gave them extra re-rolls. Ideogram’s person has no head, Midjourney’s Bigfoot is missing a foot, and Google’s version looks like a rough draft. However, if I had to pick a direction with the most promise, the winner is Midjourney.
Prompt #4 Black t-shirt with white type text that reads “Make Every Day Count”, female model, close up.
All three of these were a surprising home run on the first try. I was very impressed with multiple options to choose from. This is probably just personel preference, but I would totally wear the one in the middle from Midjourney.
Prompt # 5 A vector logo for a trendy fast food chain, “California SMASH Burger” bright colors, organic shapes
So Midjourney is out for misspelling ‘burger’ and I really like Google’s solution for it’s simplicity, but I said ‘trendy fast food chain’ so the winner is Ideogram, because it has the best potential to be a real logo.
Prompt #6 An Album cover, text reads “Hotel California” image of a desert oasis, cinematic.
While Ideogram’s version has beautiful art direction, it looks to much like a romance novel. And while I didn’t specify ‘1970s’, most of Google’s results looked like they came from that era (accurate). However, I think that the best combo of Hotel+California+Cinematic+Oasis is from Midjourney.
Prompt #7 An peaceful cow, standing next to a tractor, in front of a weathered barn, with mountains in the background, sunset.
This is the bonus round where I wrote what I thought was a very difficult prompt with multiple layers of depth to test prompt coherence. This is where Ideogram is supposed to shine, but I think they are all equally excellent, so I’m calling this one a draw.
Prompt #8 Straight from Ideograms Blog
So this might be an unfair advantage, but I used the prompt that Ideogram is currently touting in their PR campaign. Let’s see how the other two do.
A fascinating family portrait of a solid matte red sphere Christmas present perched atop a blue cube wrapped as a Christmas present, with a vivid green Christmas tree standing behind them with a green cloth wrapped around its base. A curious dog is positioned on the right and a cute cat lounges on the left.
Google really struggled with this one, and only one image was even close. Midjourney got the prompt coherence, but the image quality wasn’t great. So the winner (of their own contest) was Ideogram. With this extra help, Ideogram now ties with Midjourney with three wins each, while Google had one. A few observations: Google frequently had the fastest results, but also the greatest number of fails. I think they were censoring me, but the error message was unclear. Their were several other prompts that I didn’t include because Google produced no result. Ideogram has a built in advantage with it’s prompt helper; when I chose not to use it, the results weren’t as good. Dalle-3 has this ability as well and I will add it to the fold in my next test. In conclusion, I think that Ideogram is now one of the best tools out there and has incredible prompt coherence and text rendering ability. If you’re new to image generators, I think it’s a great value and you should check it out, because I think it’s here to stay. However, I’m sticking with Midjourney as my winner because it has a more robust feature set (upscaling, inpainting, etc) that helps in my professional workflow. Oh yeah, Midjourney 7 ships later this month, so expect another leap forward.
What do you think? Did I get it right? Which tool do you think is best? Leave a comment down below. And if you’d like more like this delivered to your inbox, sign up for my newsletter: https://erikfadiman.substack.com/
